Mapping Particle-Size Fractions as a Composition Using Additive Log-Ratio Transformation and Ancillary Data

During morphological description, one of the first properties assessed is soil texture. This is because it does not change significantly with time and it has the greatest impact on management (e.g., determining nutrient holding capacity). There is therefore an increasing need for high-resolution digital prediction of soil texture. Unfortunately, determining the three particle size fractions (PSFs) using laboratory methods is time consuming. To add value to limited soil data, ancillary data coupled with spatial and nonspatial statistical methods can be used. However, the most commonly used technique, multiple linear regression (MLR) of individual PSFs, does not consider the special requirements of a regionalized composition. We coupled ancillary data via MLR modeling to an additive log-ratio (ALR) transformation of the PSF to meet these requirements. The ancillary data included digitized air photo (e.g., Red digital numbers [DN]) and electromagnetic (EM38 and EM31) data. We found that for predicting clay at various depths, the EM38 and EM31 data were most useful. This was similarly the case for sand, with Red DN and the trend surface of some value. We also compared how prediction might be improved by using EM data measured on transects (which simulate measurements made on 1-m transects) with interpolation from transects spaced 24 m apart. The results indicate that the use of EM data on a 1-m transect using ALR-MLR can improve precision by around 24% for clay, 3% for silt, and 17% for sand with regard to topsoil prediction. We also conclude that the ALR-MLR technique has the advantage of adhering to the special requirements of a composition, with predicted values non-negative and PSFs summing to unity (i.e., = 100%).